Why it matters
A written AI policy describes what should happen. It cannot stop a customer list from being pasted into a chatbot. Runtime enforcement moves the rule to the point of use: the browser, the desktop, or the gateway an agent calls its tools through.
Because a control that misfires blocks real work, enforcement is rolled out in stages. A policy first runs in monitoring, so you see what it would have done, and a person decides when it is allowed to warn or block.
How Qastel handles it
Qastel enforces policies in the browser through its extension, on the desktop through its agent, and at the MCP gateway, with block, redact, warn, alert and quarantine effects. Every policy starts in monitoring.